Overcoming Extreme Weather Operation Challenges: The Technical Foundation Behind UISEEs 9.2 Million Kilometers of Safe Autonomous Driving Operations

2026-07-03

On July 1st, UISEE unveiled its all-weather

L4-level autonomous driving solution at MobilityTech Asia 2026 in Bangkok,

showcasing the technical foundation behind the safe operation of AI drivers in

extreme weather conditions

In autonomous driving operations, extreme weather

such as typhoons, heavy rain, snow and ice, and sandstorms are frequently

encountered, posing severe challenges to logistics transportation. The sudden

drop in visibility, sensor interference, and loss of positioning signals caused

by extreme weather represent the ultimate challenge to the limits of perception

and positioning capabilities and safety warning systems.

At these moments when "you cant see

clearly," a mature and reliable

perception warning and safety fallback mechanism is what gives AI drivers

the confidence for "truly unmanned" operations.

Why Is Extreme

Weather a "Major Test" for Autonomous Driving?

Driving in extreme weather, human drivers slow

down, increase following distance, and rely on experience to judge road

conditions. Autonomous vehicles rely on two core capabilities — perception and

positioning — to "see the environment" and "find their

position." Extreme weather negatively impacts perception and positioning

capabilities from multiple dimensions.

<img width="100%" src="/uploads/news/news-1783049621821.png" alt="">Heavy rain causes blurry camera imaging at night

When raindrops adhere to lenses, standing water

reflects light, and rain curtains reduce visibility, obstructed camera vision

leads to reduced visual recognition accuracy, increasing the risk of missed and

false detections in target detection. Water mist particles reflect laser light,

generating a large number of false point clouds, and diffuse reflection on

standing water surfaces also interferes with point cloud feature matching,

directly affecting positioning accuracy. Under severe convective weather, GPS

signals are susceptible to ionospheric disturbances and multipath effects, and

RTK positioning may experience short-term loss of lock or accuracy degradation.

Standing water covering ground markings and rain washing away road markings

further increase positioning difficulty.

<img width="100%" src="/uploads/news/news-1783049628837.png" alt="">Snowy weather generates a large number of noise points in LiDAR

When both perception and positioning capabilities

decline simultaneously, the autonomous driving system faces an exam without

"reference answers." In real B-end operation scenarios such as

airport aprons, industrial parks, and port terminals, unplanned and disorderly

shutdown of unmanned vehicles in extreme weather causes losses that quickly

propagate and amplify along the operation chain, ultimately resulting in direct production and operation losses, safety

secondary risk losses, and long-term hidden cost losses for customers.

Therefore, it is crucial for the

large-scale implementation of autonomous driving that unmanned vehicles can

make correct decisions in extreme weather, rather than operating blindly or

simply "breaking down".

Multi-Dimensional

Safety Mechanisms Overcome Extreme Weather Challenges

UISEEs U-Drive® intelligent driving system has a

core philosophy in extreme scenarios: let the AI driver act like an experienced professional driver — knowing when to slow down, when to stop, and when to

request takeover. Centered around the three goals of "uninterrupted perception, non-drifting positioning, and risk-free

decision-making", it has built multi-dimensional safety warning and

redundant protection mechanisms to ensure that unmanned vehicles remain

reliable in extreme weather.

Industry-leading multi-modal sensor fusion algorithm: Uninterrupted perception in extreme weather

UISEE has been deeply engaged in the autonomous

driving field for ten years, driven by actual business operations. In terms of

perception, it has formed a complete and mature technical system and reached

industry-leading levels, ensuring that unmanned vehicles achieve

"uninterrupted perception" in extreme weather.

In the field of multi-modal fusion perception,

UISEE has independently developed a BEV

point cloud and image fusion algorithm based on sparse paradigm. Centered

on the Transformer architecture, this algorithm efficiently fuses multi-modal

information such as images and point clouds, and enhances the feature

expression capability of the backbone network through large-scale data

pre-training, achieving industry-leading (SOTA) levels in both perception

accuracy and detection distance.

On this basis, UISEE further introduces the OCC

(Occupancy Network) module, mapping

fusion features in BEV space into dense 3D voxel occupancy predictions,

achieving centimeter-level fine depiction of the road environment (including

irregular obstacles, construction areas, drivable space boundaries, etc.),

effectively compensating for the shortcoming of traditional target detection in

perceiving non-standard obstacles.

Meanwhile, relying on the massive extreme weather

data accumulated by the company over a long period, this algorithm demonstrates

excellent robustness in harsh environments. In addition, through in-depth engineering performance

optimization, the network can maintain real-time operation on embedded

platforms.

Two-pronged multi-source fusion positioning system: Non-drifting

positioning in extreme weather

UISEE ensures "non-drifting positioning"

in extreme weather, which is corely achieved through a multi-source fusion redundant positioning system + extreme

scenario-specific optimization.

Facing extreme environments such as heavy rain,

heavy snow, and sandstorms, a single sensor often fails or its accuracy drops

sharply. UISEEs positioning module fuses

six to seven positioning sources from multiple sensors for fusion positioning,

and uses deep learning to stably handle the impact of weather changes,

enabling unmanned vehicles to achieve centimeter-level positioning in a wide

range of environments.

Laser and vision extract environmental features

from point cloud and image dimensions respectively, combined with semantic

positioning analysis of structured information such as lane lines and traffic

signs, RTK provides global coordinate reference, and wheel speedometers and

inertial navigation provide high-frequency supplementary positioning. This

ensures that vehicles can accurately drive to designated positions in complex

environments, meeting the needs of high-precision operations under special weather

conditions.

Based on the characteristics of different extreme

weather conditions, UISEE has also conducted extensive scenario-specific optimizations. In extremely cold regions,

traditional sensors experience data drift and response delays due to low

temperatures; through custom cold-resistant hardware components, sensor

performance stability is ensured at -25°C. In heavy rain and sandstorm weather,

LiDAR point clouds generate a large number of noise points; after filtering

interference through denoising algorithms, they are deeply fused with inertial

navigation and Beidou signals to ensure positioning accuracy remains

uncompromised. When snow covers ground markings and causes visual positioning

to fail, the system switches to LiDAR SLAM-dominant mode, relying on environmental

contour features to maintain positioning continuity. This two-pronged

optimization approach gives the positioning system stronger environmental

adaptability.

Industry-original

vehicle-cloud collaborative safety design: Risk-free

decision-making in extreme weather

The premise of "risk-free

decision-making" in extreme weather is to first make clear and

quantifiable judgments of risks, so that every driving decision has multiple

safeguards and is evidence-based, rather than recklessly "taking a gamble."

UISEE adopts vehicle-cloud

collaborative safety design, ensuring stable operation of L4-level unmanned

vehicles with multiple safety mechanisms. The vehicle end is the first line of

defense for decision safety; through full-stack redundancy of key components,

the possibility of single-point vehicle failures is greatly reduced, and

multi-level failure monitoring and response mechanisms ensure that vehicles can

safely park or smoothly degrade operation in the event of failures. More

importantly, the independent safety monitoring domain collects the operation

status of each module in real time and judges the vehicle safety status. Once

an abnormality in the main decision is detected, it can directly trigger safe

parking or degradation strategies. The multi-level degradation mechanism will automatically smoothly transition from

"normal driving" to "decelerated operation" and then to

"safe parking" according to the severity of weather and sensor

availability, avoiding risks caused by sudden decision changes.

The cloud end provides a global perspective and higher-dimensional judgment capabilities for decision-making. Relying on powerful computing power, the cloud has built a

situational awareness model covering nearly a thousand potential risk

scenarios, achieving comprehensive simulation and contingency handling of

various failure scenarios. Through the vehicle-cloud collaborative safety

mechanism, unmanned vehicles will not experience "one-size-fits-all"

parking when encountering extreme weather, but instead like experienced drivers, slow down when they should, stop when they

should, minimizing the impact on actual business.

Battle-Tested

Extreme Weather Safety Answers

From typhoons in Hong Kong to ice and snow in

Xinjiang, from sandstorms in the Middle East to heavy rain in North China, with

over 9.2 million kilometers of truly unmanned operation mileage, UISEEs

autonomous driving fleet has completed regular operation verification in

various extreme climates.

In 2025, Hong Kong experienced two No. 10 typhoons.

During extreme heavy rain, when Hong Kong International Airport faced

difficulties finding drivers and maintaining operations, UISEEs unmanned

vehicles still steadily moved forward through the wind and rain. By overcoming

difficult problems such as hardware instability and algorithm interference in

extreme weather, AI drivers achieve "not

selfish, not tired, not complaining, working three shifts without taking leave",

providing timely support for the airport to maintain normal operations in

extreme weather, demonstrating the precious value of technology.

In Urumqi, UISEEs unmanned vehicle fleet set a

record for large-scale commercial use of high-level autonomous driving in

extremely cold scenarios at civil aviation airports in China. Under the extreme

test of -25°C extreme cold and blizzard ice accumulation that is common in

Xinjiang winters, UISEE successfully solved

multiple industry problems such as low-temperature sensor failure, icy road

skidding, complex obstacle avoidance, and stable operation in harsh weather,

providing a scalable and replicable Chinese solution for the construction of

smart airports in extremely cold regions worldwide.

As the global climate continues to change, extreme

weather will become the "new normal" we need to face, and all-weather

operation capability will become a necessary condition for L4-level autonomous

driving. UISEE will continue to promote autonomous driving toward all-weather

and fully unmanned operation through technological innovation, so that AI

drivers can be "as steady as a rock" in any weather.